Research On Reconstruction Of Few Cycle Laser Pulse Based On Deep Neural Network

With the development of laser technology,few-cycle pulses play a huge role in the fields of physics,chemistry,biology and medicine.In the field of ultrafast optics,a few-cycle pulse is an important condition for obtaining a single attosecond pulse,which can significantly improve the time resolution of our research object,so that we can study the dynamic process of electrons on the attosecond time scale.The resulting problem is the complete measurement of few cycle pulses,the response speed of ordinary electronic components is far from enough,so researchers have proposed optical measurement methods to deal with this problem,such as frequency-resolved optical gate(FROG)and spectral phase interferometry direct electric reconstruction(SPIDER).Among them,due to the advantages of high accuracy,easy operation,low cost,and intuitive images,FROG is widely used.It is necessary to use the iterative algorithm to reconstruct the measured trace pattern during use,but the trace pattern with fewer cycle pulses is more complicated.Very many iterations are required,which greatly increases the time cost.To solve this problem,based on the work of some scientific researchers,we propose to use experimental spectral data to generate samples and reconstruct fewer laser pulses based on deep neural networks.This reconstruction method is greatly shortened compared to traditional reconstruction methods.The phase is more accurate,the anti-noise ability is better,and it is very promising.It is hoped that it will replace most of the reconstruction algorithms in the future.We successfully built a set of single-shot second harmonic FROG in the laboratory and put it into use,and built a neural network to reconstruct the FROG trace graph using deep learning.The introduction introduces the relevant background,including the generation and measurement methods of few-cycle pulses,and the advantages of deep learning.The second part is the theoretical part of a single shot FROG,including the basic principles and types of FROG,as well as the commonly used algorithm for reconstructing the trace pattern.The third part is our related research on the deep learning reconstruction of the trace graph,introducing the concepts of deep learning and the gradient descent back propagation algorithm used by the neural network,and introducing the basic structure of the convolutional neural network we use,the convolutional neural network we built,and used it to reconstruct the theoretically calculated trace graph and achieved good results.The fourth part introduces the single-shot FROG device we built and the results of the few-cycle pulses measured by the device.The fifth part is to combine the neural network we built with the single-shot FROG device,and use the neural network to reconstruct the traces measured in the experiment,and get good results.Finally,we look forward to the application of deep learning in the frontiers of physics.